About the role

Wherobots is looking for a passionate, skilled, and experienced Machine Learning Engineer to help architect, build, and operate the large-scale geospatial ML platform that powers GeoAI workflows on hundreds of terabytes to petabytes of raster data.

This is a distributed-systems-first role with meaningful ML infrastructure ownership. You will spend most of your time building high-throughput, GPU-aware data pipelines that turn massive raster archives into features, predictions, and published outputs at global scale. The role sits at the intersection of distributed systems, ML inference, and geospatial data infrastructure. If you can design clean dataflow, get the most out of a GPU cluster, and turn research prototypes into resilient production systems, we should talk.

We are 100% cloud-native and build our product using modern, reliable tooling. We use Ray, PyTorch, and the scientific Python stack (PyArrow, NumPy, Xarray) to operate on Zarr, Cloud-Optimized GeoTIFF (COG), GeoParquet, and Parquet data on object storage.

If you are passionate about building cutting-edge ML infrastructure for the physical world and want to be part of a fast-growing company at the forefront of geospatial technology, we would love to hear from you. Apply now and join the Wherobots team!

Responsibilities

• Design and operate end-to-end ML pipelines: Build pipelines over massive raster archives such as Zarr and COG, from ingestion to feature generation to inference to publication.
• Build high-throughput distributed pipelines: Use Ray (Datasets and actors) with careful control over I/O, compute overlap, and backpressure to keep clusters fully utilized.
• Optimize GPU inference at scale: Tune PyTorch inference pipelines using batching, CUDA stream overlap, and memory-aware scheduling to maximize throughput per GPU.
• Develop spatial data processing patterns: Implement tiling, overlapping windows, and accumulators that match the access patterns of spatial models.
• Ensure production reliability: Build in retries, checkpointing, observability, and cost-efficient scaling so long-running global jobs are debuggable and resilient to failure.
• Build reusable platform abstractions: Collaborate on abstractions that generalize across datasets, models, and product use cases so new workflows ship quickly.
• Raise the bar: Provide technical leadership on architecture, engineering standards, and roadmap, and contribute to architecture and code reviews across the organization.

Qualifications

• 5+ years of experience building distributed data or ML systems in production.
• Strong hands-on experience with Ray, Spark, Dask, or similar distributed computing systems.
• Deep understanding of performance tradeoffs across memory, I/O, serialization, and scheduling.
• Production experience running GPU inference workloads at scale.
• Proficiency in Python and the scientific Python stack (PyTorch, PyArrow, NumPy, Xarray).
• Strong working knowledge of object storage (e.g., S3) and large-scale data access patterns such as range requests and sharding.
• Experience with columnar storage formats (GeoParquet, Parquet) and chunked storage formats (Zarr v3).
• Geospatial experience, especially with raster data, projections, or tiling schemes.

Nice to Have (Optional)

• Experience designing data platforms or reusable ML infrastructure.
• Familiarity with the lakehouse architecture and table formats
• Experience turning research or prototype ML workflows into production systems.

Compensation and benefits

Wherobots offers competitive compensation, equity, and benefits. The base salary range for this position is $185k-$275k per year.

Preferred locations: San Francisco Bay Area or Seattle. We provide flexibility in working arrangements for most roles, including remote, hybrid, and in-office options. For candidates who receive an offer, base pay varies based on location, seniority, skills, and experience.

Our benefits package for full-time employees includes:

• 100% coverage of medical, dental, and vision insurance.
• Access to a 401(k) retirement plan.
• Unlimited paid time off.

About Wherobots

Wherobots is the AI Context Engine for the Physical World: the missing infrastructure layer for AI that needs to reason about our physical reality. Existing AI context infrastructure was built for text, code, documents, databases, and the internet. The most consequential enterprise decisions also depend on the physical world: supply chains, climate exposure, geopolitics, infrastructure, and more.

Our two products let teams use familiar SQL and Python to analyze physical-world data and give AI agents persistent context about their organization’s assets and operations:

• WherobotsDB is a fully distributed spatial analytics engine with a Rust-native execution layer. It processes vector and raster data in the same engine at scale, enabling precise spatial joins across billions of geometries and analysis that connects imagery with business data.
• RasterFlow is a fully distributed, serverless engine for raster preparation and computer vision. It processes satellite imagery at planetary scale, runs built-in or custom PyTorch models, and turns imagery into predictions and spatial features that teams can analyze with WherobotsDB.

Wherobots was founded by the original creators of Apache Sedona, the open-source spatial computing platform with more than 80 million downloads. Apache Sedona is used in production by tens of thousands of organizations, including many Fortune 500 companies.